How the shift runs

They watch, they decide,
then they act.

Most tools stop at the first job. Everything that makes a difference happens in the second and the third.

What
What is QuikSignal?
A team of AI agents that researches everything about the companies you choose - their news, their job ads, their filings, their code, their prices - and tells you the few things that change a decision. It runs every night, in your own workspace, and every claim it makes carries the document it came from.
How
How does it work?
You name your market and the companies in it. From then on the agents research it on your behalf every night, throw away the ninety-nine per cent that is noise, and hand you what is left: what happened, which of your accounts it happened to, who holds the decision, and what to do about it this week.
Why
Why would you want it?
Because the thing that would have won you the deal was on the record, in public, six weeks before you heard about it. Somebody has to read the market every day for that to be worth anything - and nobody has the day. See it on your own market.

On shift all night, every night

Newsrooms, filings, job boards, funding databases, earnings calls, pricing pages. Nobody calls in sick, nobody forgets, nobody stops at five.

The shift starts

Almost all of it is noise

A thousand things happen. Seven change a decision. Your team keeps those seven, and always shows you why it kept them.

The shift decides

It arrives where you already are

The brief in your inbox before your first meeting. The account, the reason, and the person to call - with the source document that says so.

The shift delivers

How the shift decides

A thousand things happened. They kept seven.

Almost none of this needs a model. Deduplication, clustering and scoring are deterministic, which is why the whole night costs so little and finishes before you wake up.

One night, in four steps

Deduplication, clustering and scoring are deterministic - no model sees 99% of this.

Collected1,522

Your whole market, researched in full.

Deduplicated8% of what came in120

The same event told twelve ways becomes one story.

Shortlisted1% of what came in20

Scored on trigger strength, source weight and recency.

Sent0.5% of what came in7

Cross-checked against a second source, then written up.

Bar length is linear, so each step is drawn in true proportion to the one above it. Seven out of 1,412 is half a percent, and it looks like it.

Demo data from a representative workspace.

Key takeaway

Last week this workspace also collected 1,405 other things - and held every one of them back. You can read the list of what it withheld, and why.

The same night, measured

The hard part is deciding what to throw away.

Fifteen hundred documents in a night; four that change a decision. Researching a market is not the hard part - a machine can read anything. Here the throwing away is done by rules you can read rather than by a model you have to trust.

One night, one workspaceMeasured on a real run, not modelled. Ninety-nine per cent never reaches a model at all - which is what makes reading this much affordable.

Read0Documents pulled from thirty live sources
Clustered0Near-duplicate coverage folded into one story each
Scored0Cleared the trigger floor and went to a model
Delivered0Changed a decision, and reached a human

Where the evidence comes fromAll 30 of them live today, keyless and public every one - so nothing here depends on a subscription you have to keep renewing.

News & web5
Company intelligence7
Technology7
Hiring6
Filings & registers3
Contracts & awards2

Key takeaway

The reduction is deterministic: near-duplicate coverage is clustered, ten weighted trigger patterns are scored, recency decays over twenty-one days, and only what clears the floor is sent to a model. That is why the same night produces the same shortlist twice.

From a signal to an action

Thousands of documents in.
One call out.

Companies put things on the record every day - a round, a role, a price, a filing. We research all of it, keep the few that change a decision, work out who holds that decision, and hand you six messages written from the evidence.

What is on the record, and whereworked example

0documents read last night-changed a decision

What happenedBrightline Payroll raised $40m

Series B, led by Northwind. Three independent sources, and the filing itself.

Who to call, and why this weekTheir new Head of Data

Named in the funding release. They have money, a mandate and no delivery partner - and they are hiring the build layer without the lead layer.

Six messages you would actually send

“You have just closed a round and opened three platform roles without a principal - that order usually means the architecture is still open.”

1Something happensA company raises money, opens a role, changes its price, files with a regulator. In public, with a date on it.
2We research itEvery document about the companies you watch - thousands a week - and we keep the handful that changes a decision.
3We work out who, and why nowThe person who holds that decision, named in a document we can show you, with the reason to call them this week.
4You get six messagesWritten from the evidence, per person, checked and scored before you see them. You send them, or you do not.

The same thing, stage by stage

Six stages, each one a thing you can watch happen in your own workspace on the first day. The sixth is the one worth knowing about: it scores the writing and sends it back to be done again when it falls short.

Stage 1 of 6

Something changed

A round closed, a price page moved, a role opened, a filing landed. Found in the evidence, with the document that says so.

SEC filing · 8-KPress releaseTrade press

Brightline PayrollSeries B · $40m · 3 independent sources

What it producesBrightline Payroll · Series B · 3 sources

The same account, read for two different sellers

The join is what makes this yours rather than generic: your seller profile is drafted from your own website and approved by you before a single line is written on your behalf. Two companies looking at the same account get two different answers.

Selling for A data engineering firm read from their site

What the account is missingNo feature storeBatch-only risk scoringManual SOC2 evidenceEleven open roles, none seniornothing they sell answers this

What you sellStreaming data platformsMLOps and governanceCloud and FinOpsIntelligent automation

Same account, same evidence, two different sellers. The gaps do not move - what connects to them does. A gap with no line is the useful part: it is the week you do not spend on an account nothing you sell fits.

Key takeaway

Quality is checked deterministically, not by asking another model whether it did well. Generic phrasing, claims that reference nothing in the evidence, and near-duplicates across people are each detected and counted. Below 75 out of 100 it rewrites itself once, keeps whichever attempt scored better, and shows you the score either way.

The night shift

You wake up to one message, not a to-do list.

A filing went up at two in the morning. By breakfast it had been matched, checked, written up and filed against the right account. Nobody touched it.